ASCENT: Adaptive Self-Configuring sEnsor Networks Topologies. Alberto Cerpa and Deborah Estrin, In Proceedings of the Twenty First International Annual Joint Conference of the IEEE Computer and Communications Societies (INFOCOM 2002), New York, NY, USA, June 23-27 2002. November 21, 2018 Haeyong Kim (hykim@mmlab.snu.ac.kr) Seoul National Univ.

Distributed sensor network scenario Ex) A habitat monitoring sensor network Deployed in a remote forest by dropping sensor nodes from a plane or by placing them by hand Ad-hoc deployment Cannot expect the sensor field to be deployed in a regular fashion Uniform deployment does not correspond to uniform connectivity Energy constraints Expend as little energy as possible to maximize network lifetime Unattended operation under dynamics Preclude manual configuration, pre-configuration Haeyong Kim (hykim@mmlab.snu.ac.kr) Seoul National Univ.

Assumptions Reacts when links experience high packet loss High enough node density to connect the entire region When node density is low, ASCENT mechanism does not applicable (generally, all nodes should be used to form an effective network) Does not consider partition (partition detection and repair techniques are leaved to future work) Haeyong Kim (hykim@mmlab.snu.ac.kr) Seoul National Univ.

Primary contributions Use of adaptive techniques Applications configure the underlying topology based on their needs while trying to save energy to extend network lifetime Use of self-configuring techniques React to operating conditions measured locally Use of adaptive techniques : Does not presume fairness, degree of connectivity or capacity required Haeyong Kim (hykim@mmlab.snu.ac.kr) Seoul National Univ.

ASCENT design ASCENT adaptively elects “active” nodes “passive” nodes Stay awake all the time and perform multi-hop packet routing “passive” nodes Periodically check if they should become active Two-hop network example Only some nodes are active. Very high message loss (because range) sending “help messages” passive neighbor nodes join the network Neighbor receive help message, it may decide to join the network (becomes an active neighbor) signals the existence of new active neighbor to other passive neighbors by sending a “neighbor announcement message” continues until the number of active nodes stabilizes on a certain value Delivery of data from source to sink more reliable Haeyong Kim (hykim@mmlab.snu.ac.kr) Seoul National Univ.

ASCENT state transitions Random timer turns on the nodes to avoid synchronization When a node starts, test state Exchange data and routing control messages and sets up a timer (Tt), sends neighbor announcement messages with node ID Before Tt expires, neighbors > NT (neighbor threshold) or DL (data loss rate) > DL before entering in the test state passive state Passive  test : active neighbor가 보낸 help 메시지. LT (loss threshold) Passive mode : overhear all packet (even if not addressed to the passive node)  gather information Haeyong Kim (hykim@mmlab.snu.ac.kr) Seoul National Univ.

ASCENT parameter tuning Choices left to the applications NT (neighbor threshold) the average degree of connectivity of the network Trade-off between the energy consumed and/or the level of interference vs. the desired sensing coverage LT (loss threshold) The maximum amount of data loss that an application can tolerate Application dependent temperature measurements vs tracking of a moving target Tt, Tp (timer of test state, passive state) Trade-off of power consumption vs. decision quality Ts (timer of sleep state) Trade-off of energy saving vs. reaction time to dynamics Level of interference = packet loss Tt, Tp : Robust decision in presence of transient paket losses, neighbor determination에 영향 Haeyong Kim (hykim@mmlab.snu.ac.kr) Seoul National Univ.

Neighbor and data loss determination Measured locally by each node while in passive and test state Each node increase the sequence number when each packet transmitted When a sequence number is skipped, loss is detected Assumes application data packets also have some mechanism to detect losses N이 커질수록 loss가 커지게 되므로, loss threshold도 같이 증가 시켜야만 합리적인 N값을 얻을 수 있다. 그렇지 않으면 active neighbor가 증가함에도 불구하고 N값은 줄어들 수가 있다. Haeyong Kim (hykim@mmlab.snu.ac.kr) Seoul National Univ.

Neighbor and data loss determination (cont’d) The number of active neighbors (N) The number of neighbors with link packet loss smaller than the neighbor loss threshold (NLS) NLS = 1- (1/N) This formula worked best N : the number of neighbors calculated in the previous cycle NLS : neighbor loss threshold Average data loss rate (DL) Calculated based on the application data packets Data losses are detected using data sequence numbers If the message was not received from any neighbor, it is considered a data loss Control messages are not considered Help, neighbor announcement and routing control Depending on the routing strategy, a node may receive multiple copies of the same application data packet. Haeyong Kim (hykim@mmlab.snu.ac.kr) Seoul National Univ.

ASCENT interactions with routing runs above link and MAC layer below routing layer is not a routing or data dissemination protocol decides which nodes should join the routing infrastructure Nodes become active or passive independent of routing protocol Does not use state gathered by the routing protocol Does not requires changing the routing state Test state (actively routing packets)  passive state (listen-only) Cause some packet loss Improvement : Traffic could be rerouted in advance by informing the routing protocol of ASCENT’s state changes Haeyong Kim (hykim@mmlab.snu.ac.kr) Seoul National Univ.

Conclusion ASCENT has the potential of ASCENT mechanisms were significant reduction of packet loss Increase in energy efficiency ASCENT mechanisms were responsive and stable under systematically varied conditions Future work Use of load balancing techniques to distribute the energy load Use of wider area links to detect network partitions Haeyong Kim (hykim@mmlab.snu.ac.kr) Seoul National Univ.